<
competition-corner-p...@googlegroups.com> writes:
> [ . . . ] Problem suggested by David McCooey
> [ . . . ]
> Find coordinates for a regular heptagon in 3D Euclidean space where
> all 3 components (x,y,z) of all 7 coordinates are elements of the same
> cubic field, or prove that it can't be done.
That's a nice problem, which I haven't seen before.
It looks like such a heptagon is impossible. My proof follows.
It is more convenient to work with the algebraic integer
d = 2 c = 2 cos(2 Pi/7) = z + 1/z
where z is a 7th root of unity; the minimal equation of d is
d^3 + d^2 - 2*d - 1 = 0. David Ash already showed that
if such a heptagon exists then the field F must be QQ(c),
which is also QQ(d), and that 1-c^2 (equivalently,
4(1-c^2) = 4-d^2) must be a sum of three squares in F.
This turns out to be impossible, for much the same reason that
7 is not a sum of three rational squares: Check that if
X,Y,Z are algebraic integers in F then X^2 + Y^2 + Z^2
(i) cannot be -w^2 mod 8 for any algebraic integer w
that's not a multiple of 2, and (ii) can only be a multiple of 4 if
each of X,Y,Z is a multiple of 2. Then if X^2 + Y^2 + Z^2 = 7 T^2
for some algebraic integers X,Y,Z,T in F with T nonzero
then we get a contradiction by repeated applying (ii) until
finding an equivalent solution with T not divisible by 2,
at which point (i) shows that this is impossible because
4-d^2 is congruent to -(2d^2-d+2)^2 mod 8. [We say that
two algebraic numbers are "congruent mod m" iff they differ by
c*m for some _algebraic_ integer c.]
Some further remarks:
@ David Ash outlined a proof of the equivalence
"heptagon exists <==> 1-c^2 = X^2 + Y^2 + Z^2 in F",
but since we need only the "==>" direction
a simpler proof suffices. Call the cubic field K.
The center of the heptagon has coordinates in K
(it's the average of the coordinates of the vertices),
so we might as well translate to put the center at the origin.
Then all the vertices have the same length, say sqrt(N)
for some nonzero N in K. If v and v' are adjacent vertices
then the dot product v.v' is c N; since v.v' and N are both in K,
so is c, whence K = QQ(c) = F. The cross product of v and v'
is a vector of length s / N; thus s^2 / N^2 = (4-d^2) / (2N)^2
is a sum of three squares in F, whence so is 4-d^2.
@ It so happens that (4-d^2) (d^2-d-2)^2 = 7,
so that (i) above is even more clearly a generalization of
the familiar fact that 7 is not the sum of three rational squares.
@ I checked (i) and (ii) by having the computer run through
all possible of residue classes in ZZ[c] mod 4 or 8.
(First get a list of all possible squares mod 8;
then sort, remove duplicates, and loop over triples in
the resulting duplicate-free list.) I think that I can
give a proof that generalizes this to all fields QQ(x)
generated by a root of some odd-degree polynomial in ZZ[x]
that remains irreducible mod 2 -- but this may be beyond
the scope of the competition-corner-participant list.
@ A regular 9-gon *can* be placed in Euclidean space
with all coordinates in the same cubic field, necessarily
the field generated by 2*cos(2*Pi/9), which is a root of
t^3 - 3*t + 1 = 0. Indeed we may put three vertices at
an equilateral triangle such as (0,1,-1), (-1,0,1), (1,-1,0),
then fill in the remaining 6 vertices as QQ[t]-linear combinations.
Here if we imitate our analysis for the heptagon we end up with
4-t^2 being a sum of three squares, which it is because
4-t^2 = (2t^2+t-4) / 3.